GPT-5.6 gets the green light as ChatGPT Work pushes Codex beyond developers

OpenAI is ending the limited preview of GPT-5.6, which The Verge says had been available only to government approved organizations, and is moving the model into public rollout. The company also introduced ChatGPT Work, an agent powered by the GPT-5.6 model suite: Sol, Terra and Luna.

GPT-5.6 moves from a government filtered preview to public rollout

Sam Altman called GPT-5.6 the best model OpenAI has ever produced. The more useful detail is access. According to The Verge, Mac and Windows users worldwide, including free ChatGPT users, should get ChatGPT Work and GPT-5.6 immediately through the desktop app.

On mobile and web, the sequence is more cautious: Pro, Enterprise and Edu first, then Plus and Business over the next few days. The rollout is described as global and moving toward full availability over 24 hours.

ChatGPT Work turns Codex from a developer tool into an office agent

ChatGPT Work is described as a combination of ChatGPT and Codex for non technical users. It can gather context from chosen apps, files and workflows, then create documents, spreadsheets, presentations and web apps.

That product move matters more than the model version. OpenAI is taking the agent story from code generation into the work currently handled through people, templates, Slack, Gmail, Google Drive, calendars and CRMs. The unified plugins directory is the connective layer.

The office agent will hit permissions before it hits intelligence

The weak point is not just output quality. Once an agent reads email, files, calendars and CRM records, the hard problems become permissions, audit and the line between „help me“ and „make the decision for me“.

OpenAI is also framing Sol as strong in coding, cybersecurity, science and computer use. Without independent evals and clear pricing details, claims about being a lower cost alternative to competitors remain OpenAI’s marketing frame rather than a verified comparison.

Connector governance will matter more than the demo document

The next signal is whether companies allow ChatGPT Work into sensitive systems or confine it to a sandbox. An agent without access is a pricier editor. An agent with access is a new security surface.

The enterprise details to watch are admin controls: connector approvals, logs, exports, roles and the ability to disable specific capabilities. That is where ChatGPT Work either becomes daily infrastructure or stays a large demo inside the desktop app.

Lilith's verdict

ChatGPT Work is a salesperson holding a keyring for company apps. If firms clip Gmail, Drive and CRM onto that ring, the model stops waiting in chat and starts standing at the office door.